Clean up SkXfermode_opts.h

src/opts/SkXfermode_opts.h

 /*
  * Copyright 2015 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #ifndef Sk4pxXfermode_DEFINED
 #define Sk4pxXfermode_DEFINED
 
 #include "Sk4px.h"
 #include "SkNx.h"
 #include "SkXfermode_proccoeff.h"
 
 namespace {
 
 // Most xfermodes can be done most efficiently 4 pixels at a time in 8 or 16-bit fixed point.
-#define XFERMODE(Name) static Sk4px SK_VECTORCALL Name(Sk4px s, Sk4px d)
+#define XFERMODE(Xfermode) \
+    struct Xfermode { Sk4px operator()(const Sk4px&, const Sk4px&) const; }; \
+    inline Sk4px Xfermode::operator()(const Sk4px& s, const Sk4px& d) const
 
 XFERMODE(Clear) { return Sk4px::DupPMColor(0); }
 XFERMODE(Src)   { return s; }
 XFERMODE(Dst)   { return d; }
 XFERMODE(SrcIn)   { return     s.approxMulDiv255(d.alphas()      ); }
 XFERMODE(SrcOut)  { return     s.approxMulDiv255(d.alphas().inv()); }
 XFERMODE(SrcOver) { return s + d.approxMulDiv255(s.alphas().inv()); }
-XFERMODE(DstIn)   { return SrcIn  (d,s); }
-XFERMODE(DstOut)  { return SrcOut (d,s); }
-XFERMODE(DstOver) { return SrcOver(d,s); }
+XFERMODE(DstIn)   { return SrcIn  ()(d,s); }
+XFERMODE(DstOut)  { return SrcOut ()(d,s); }
+XFERMODE(DstOver) { return SrcOver()(d,s); }
 
 // [ S * Da + (1 - Sa) * D]
 XFERMODE(SrcATop) { return (s * d.alphas() + d * s.alphas().inv()).div255(); }
-XFERMODE(DstATop) { return SrcATop(d,s); }
+XFERMODE(DstATop) { return SrcATop()(d,s); }
 //[ S * (1 - Da) + (1 - Sa) * D ]
 XFERMODE(Xor) { return (s * d.alphas().inv() + d * s.alphas().inv()).div255(); }
 // [S + D ]
 XFERMODE(Plus) { return s.saturatedAdd(d); }
 // [S * D ]
 XFERMODE(Modulate) { return s.approxMulDiv255(d); }
 // [S + D - S * D]
 XFERMODE(Screen) {
     // Doing the math as S + (1-S)*D or S + (D - S*D) means the add and subtract can be done
     // in 8-bit space without overflow.  S + (1-S)*D is a touch faster because inv() is cheap.
@@ skipping 29 matching lines @@
     auto isLite = ((sa-s) < s).widenLoHi();
 
     auto lite = sa*da - ((da-d)*(sa-s) << 1),
          dark = s*d << 1,
          both = s*da.inv() + d*sa.inv();
 
     auto alphas = srcover;
     auto colors = (both + isLite.thenElse(lite, dark)).div255();
     return alphas.zeroColors() + colors.zeroAlphas();
 }
-XFERMODE(Overlay) { return HardLight(d,s); }
+XFERMODE(Overlay) { return HardLight()(d,s); }
 
 XFERMODE(Darken) {
     auto sa = s.alphas(),
          da = d.alphas();
 
     auto sda = (s*da).div255(),
          dsa = (d*sa).div255();
 
     auto srcover = s + (d * sa.inv()).div255(),
          dstover = d + (s * da.inv()).div255();
@@ skipping 10 matching lines @@
 
     auto srcover = s + (d * sa.inv()).div255(),
          dstover = d + (s * da.inv()).div255();
     auto alphas = srcover,
          colors = (dsa < sda).thenElse(srcover, dstover);
     return alphas.zeroColors() + colors.zeroAlphas();
 }
 #undef XFERMODE
 
 // Some xfermodes use math like divide or sqrt that's best done in floats 1 pixel at a time.
-#define XFERMODE(Name) static Sk4f SK_VECTORCALL Name(Sk4f d, Sk4f s)
+#define XFERMODE(Xfermode) \
+    struct Xfermode { Sk4f operator()(const Sk4f&, const Sk4f&) const; }; \
+    inline Sk4f Xfermode::operator()(const Sk4f& d, const Sk4f& s) const
 
 static inline Sk4f a_rgb(const Sk4f& a, const Sk4f& rgb) {
     static_assert(SK_A32_SHIFT == 24, "");
     return a * Sk4f(0,0,0,1) + rgb * Sk4f(1,1,1,0);
 }
 static inline Sk4f alphas(const Sk4f& f) {
     return SkNx_dup<SK_A32_SHIFT/8>(f);
 }
 
 XFERMODE(ColorDodge) {
@@ skipping 50 matching lines @@
 
     auto alpha  = s + d*isa;
     auto colors = s*ida + d*isa + (s2 <= sa).thenElse(darkSrc, liteSrc);           // Case 1 or 2/3?
 
     return a_rgb(alpha, colors);
 }
 #undef XFERMODE
 
 // A reasonable fallback mode for doing AA is to simply apply the transfermode first,
 // then linearly interpolate the AA.
-template <Sk4px (SK_VECTORCALL *Mode)(Sk4px, Sk4px)>
-static Sk4px SK_VECTORCALL xfer_aa(Sk4px s, Sk4px d, Sk4px aa) {
-    Sk4px bw = Mode(s, d);
+template <typename Xfermode>
+static Sk4px xfer_aa(const Sk4px& s, const Sk4px& d, const Sk4px& aa) {
+    Sk4px bw = Xfermode()(s, d);
     return (bw * aa + d * aa.inv()).div255();
 }
 
 // For some transfermodes we specialize AA, either for correctness or performance.
-#define XFERMODE_AA(Name) \
-    template <> Sk4px SK_VECTORCALL xfer_aa<Name>(Sk4px s, Sk4px d, Sk4px aa)
+#define XFERMODE_AA(Xfermode) \
+    template <> Sk4px xfer_aa<Xfermode>(const Sk4px& s, const Sk4px& d, const Sk4px& aa)
 
 // Plus' clamp needs to happen after AA.  skia:3852
 XFERMODE_AA(Plus) {  // [ clamp( (1-AA)D + (AA)(S+D) ) == clamp(D + AA*S) ]
     return d.saturatedAdd(s.approxMulDiv255(aa));
 }
 
 #undef XFERMODE_AA
 
+template <typename Xfermode>
 class Sk4pxXfermode : public SkProcCoeffXfermode {
 public:
-    typedef Sk4px (SK_VECTORCALL *Proc4)(Sk4px, Sk4px);
-    typedef Sk4px (SK_VECTORCALL *AAProc4)(Sk4px, Sk4px, Sk4px);
-
-    Sk4pxXfermode(const ProcCoeff& rec, SkXfermode::Mode mode, Proc4 proc4, AAProc4 aaproc4)
-        : INHERITED(rec, mode)
-        , fProc4(proc4)
-        , fAAProc4(aaproc4) {}
+    Sk4pxXfermode(const ProcCoeff& rec, SkXfermode::Mode mode)
+        : INHERITED(rec, mode) {}
 
     void xfer32(SkPMColor dst[], const SkPMColor src[], int n, const SkAlpha aa[]) const override {
         if (nullptr == aa) {
             Sk4px::MapDstSrc(n, dst, src, [&](const Sk4px& dst4, const Sk4px& src4) {
-                return fProc4(src4, dst4);
+                return Xfermode()(src4, dst4);
             });
         } else {
             Sk4px::MapDstSrcAlpha(n, dst, src, aa,
                     [&](const Sk4px& dst4, const Sk4px& src4, const Sk4px& alpha) {
-                return fAAProc4(src4, dst4, alpha);
-            });
+                return xfer_aa<Xfermode>(src4, dst4, alpha);
+           });
         }
     }
 
     void xfer16(uint16_t dst[], const SkPMColor src[], int n, const SkAlpha aa[]) const override {
-        if (nullptr == aa) {
-            Sk4px::MapDstSrc(n, dst, src, [&](const Sk4px& dst4, const Sk4px& src4) {
-                return fProc4(src4, dst4);
-            });
-        } else {
-            Sk4px::MapDstSrcAlpha(n, dst, src, aa,
-                    [&](const Sk4px& dst4, const Sk4px& src4, const Sk4px& alpha) {
-                return fAAProc4(src4, dst4, alpha);
-            });
+        SkPMColor dst32[4];
+        while (n >= 4) {
+            dst32[0] = SkPixel16ToPixel32(dst[0]);
+            dst32[1] = SkPixel16ToPixel32(dst[1]);
+            dst32[2] = SkPixel16ToPixel32(dst[2]);
+            dst32[3] = SkPixel16ToPixel32(dst[3]);
+
+            this->xfer32(dst32, src, 4, aa);
+
+            dst[0] = SkPixel32ToPixel16(dst32[0]);
+            dst[1] = SkPixel32ToPixel16(dst32[1]);
+            dst[2] = SkPixel32ToPixel16(dst32[2]);
+            dst[3] = SkPixel32ToPixel16(dst32[3]);
+
+            dst += 4;
+            src += 4;
+            aa  += aa ? 4 : 0;
+            n -= 4;
+        }
+        while (n) {
+            SkPMColor dst32 = SkPixel16ToPixel32(*dst);
+            this->xfer32(&dst32, src, 1, aa);
+            *dst = SkPixel32ToPixel16(dst32);
+
+            dst += 1;
+            src += 1;
+            aa  += aa ? 1 : 0;
+            n   -= 1;
         }
     }
 
 private:
-    Proc4 fProc4;
-    AAProc4 fAAProc4;
     typedef SkProcCoeffXfermode INHERITED;
 };
 
+template <typename Xfermode>
 class Sk4fXfermode : public SkProcCoeffXfermode {
 public:
-    typedef Sk4f (SK_VECTORCALL *ProcF)(Sk4f, Sk4f);
-    Sk4fXfermode(const ProcCoeff& rec, SkXfermode::Mode mode, ProcF procf)
-        : INHERITED(rec, mode)
-        , fProcF(procf) {}
+    Sk4fXfermode(const ProcCoeff& rec, SkXfermode::Mode mode)
+        : INHERITED(rec, mode) {}
 
     void xfer32(SkPMColor dst[], const SkPMColor src[], int n, const SkAlpha aa[]) const override {
         for (int i = 0; i < n; i++) {
-            dst[i] = aa ? this->xfer32(dst[i], src[i], aa[i])
-                        : this->xfer32(dst[i], src[i]);
+            dst[i] = Xfer32_1(dst[i], src[i], aa ? aa+i : nullptr);
         }
     }
 
     void xfer16(uint16_t dst[], const SkPMColor src[], int n, const SkAlpha aa[]) const override {
         for (int i = 0; i < n; i++) {
             SkPMColor dst32 = SkPixel16ToPixel32(dst[i]);
-            dst32 = aa ? this->xfer32(dst32, src[i], aa[i])
-                       : this->xfer32(dst32, src[i]);
+            dst32 = Xfer32_1(dst32, src[i], aa ? aa+i : nullptr);
             dst[i] = SkPixel32ToPixel16(dst32);
         }
     }
 
 private:
+    static SkPMColor Xfer32_1(SkPMColor dst, const SkPMColor src, const SkAlpha* aa) {
+        Sk4f d = Load(dst),
+             s = Load(src),
+             b = Xfermode()(d, s);
+        if (aa) {
+            Sk4f a = Sk4f(*aa) * Sk4f(1.0f/255);
+            b = b*a + d*(Sk4f(1)-a);
+        }
+        return Round(b);
+    }
+
     static Sk4f Load(SkPMColor c) {
         return SkNx_cast<float>(Sk4b::Load((uint8_t*)&c)) * Sk4f(1.0f/255);
     }
+
     static SkPMColor Round(const Sk4f& f) {
         SkPMColor c;
         SkNx_cast<uint8_t>(f * Sk4f(255) + Sk4f(0.5f)).store((uint8_t*)&c);
         return c;
     }
-    inline SkPMColor xfer32(SkPMColor dst, SkPMColor src) const {
-        return Round(fProcF(Load(dst), Load(src)));
-    }
 
-    inline SkPMColor xfer32(SkPMColor dst, SkPMColor src, SkAlpha aa) const {
-        Sk4f s(Load(src)),
-             d(Load(dst)),
-             b(fProcF(d,s));
-        // We do aa in full float precision before going back down to bytes, because we can!
-        Sk4f a = Sk4f(aa) * Sk4f(1.0f/255);
-        b = b*a + d*(Sk4f(1)-a);
-        return Round(b);
-    }
-
-    ProcF fProcF;
     typedef SkProcCoeffXfermode INHERITED;
 };
 
 } // namespace
 
 namespace SK_OPTS_NS {
 
 static SkXfermode* create_xfermode(const ProcCoeff& rec, SkXfermode::Mode mode) {
     switch (mode) {
-#define CASE(Mode) \
-    case SkXfermode::k##Mode##_Mode: return new Sk4pxXfermode(rec, mode, &Mode, &xfer_aa<Mode>)
+#define CASE(Xfermode) \
+    case SkXfermode::k##Xfermode##_Mode: return new Sk4pxXfermode<Xfermode>(rec, mode)
         CASE(Clear);
         CASE(Src);
         CASE(Dst);
         CASE(SrcOver);
         CASE(DstOver);
         CASE(SrcIn);
         CASE(DstIn);
         CASE(SrcOut);
         CASE(DstOut);
         CASE(SrcATop);
         CASE(DstATop);
         CASE(Xor);
         CASE(Plus);
         CASE(Modulate);
         CASE(Screen);
         CASE(Multiply);
         CASE(Difference);
         CASE(Exclusion);
         CASE(HardLight);
         CASE(Overlay);
         CASE(Darken);
         CASE(Lighten);
     #undef CASE
 
-#define CASE(Mode) \
-    case SkXfermode::k##Mode##_Mode: return new Sk4fXfermode(rec, mode, &Mode)
+#define CASE(Xfermode) \
+    case SkXfermode::k##Xfermode##_Mode: return new Sk4fXfermode<Xfermode>(rec, mode)
         CASE(ColorDodge);
         CASE(ColorBurn);
         CASE(SoftLight);
     #undef CASE
 
         default: break;
     }
     return nullptr;
 }
 
 } // namespace SK_OPTS_NS
 
 #endif//Sk4pxXfermode_DEFINED